An intelligent monitoring and multi-robot coordination system based on hardware simulation of agricultural unmanned aerial vehicles

Abstract

This invention belongs to the field of unmanned aerial vehicle (UAV) technology, and specifically relates to an intelligent monitoring and multi-UAV collaborative system for agricultural UAVs based on hardware simulation. It achieves high-precision anomaly detection through a hybrid model of 3+ lightweight LSTM and isolated forest, with an average detection rate of 97.6% and a false alarm rate of 1.52%. A realistic physical link verification environment is constructed using Arduino hardware-in-the-loop simulation. It supports unified access for multiple UAV brands, multi-UAV collision warnings, and signal interference warnings. It adopts a lightweight web architecture, allowing for immediate use in a browser. This invention effectively solves the problems of closed ecosystems, high false alarm rates, field failures, and lack of multi-UAV collaboration in traditional solutions. It is suitable for scenarios such as plant protection operation safety monitoring, large-scale base management, and UAV teaching and training, providing highly reliable safety guarantees for low-altitude economy and smart agriculture.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of unmanned aerial vehicle (UAV) technology, and particularly relates to an intelligent monitoring and multi-UAV collaborative system for agricultural UAVs based on hardware simulation. Background Technology

[0002] The application of agricultural drones in plant protection, sowing, and inspection is growing exponentially. However, drone operations in agricultural settings face multiple challenges: Regarding weather factors, severe weather such as strong winds and heavy fog can lead to flight loss of control and a narrowing of the operating window. When wind speeds increase, spray droplets are prone to drift, resulting in uneven pesticide distribution and severely impacting spraying accuracy and pesticide utilization. Regarding terrain factors, complex terrain in hilly and mountainous areas can easily cause signal blockage, leading to a loss of communication between the drone and the control unit, a problem particularly prominent in hilly areas of southern China. Regarding equipment factors, traditional detection models are mostly built based on ideal experimental environments. In actual farmland scenarios, they are easily affected by changes in lighting, occlusion interference, complex backgrounds, and weather conditions, leading to decreased detection performance and difficulty meeting the high reliability requirements of complex agricultural environments. Furthermore, practical problems such as equipment corrosion, under-spraying width, and decreased sensor accuracy also exist.

[0003] Furthermore, traditional drone operation monitoring methods typically rely on pilot visual observation or post-operation data analysis, making it difficult to achieve real-time and intuitive risk perception. Users need a visualization platform that can display drone flight trajectories, status parameters, and risk points in real time, enabling timely detection of anomalies and prompt action during operations.

[0004] Even more serious is the fact that, with the expanding scale of agricultural drone applications, multi-drone collaborative operations have become the norm. Multiple drones operating simultaneously in the same airspace not only pose a risk of collision due to proximity but also risk communication quality degradation or even loss of connection due to interference from co-channel signals. Existing monitoring methods are mostly based on independent monitoring of individual drones, lacking unified cross-device data integration and real-time visualization capabilities. Operators often need to frequently switch between multiple systems, making it difficult to promptly detect and address safety hazards. According to the "2024 Agricultural Drone Industry Safety White Paper," the drone accident rate due to signal anomalies, flight loss of control, and other reasons is increasing year by year, causing not only huge economic losses to equipment but also threatening the personal safety of field workers.

[0005] Currently, monitoring solutions for the safety of agricultural drone operations are mainly divided into two categories, but both have obvious limitations in dealing with complex agricultural scenarios and the need for multi-drone collaboration.

[0006] (1) The closed ecosystem of leading manufacturers (the problem of data silos is prominent) Drone manufacturers such as DJI and XAG typically equip their drones with built-in flight monitoring systems. While these systems offer high levels of hardware and software integration and stable basic flight control, they generally employ closed underlying communication protocols and proprietary data formats, resulting in severe data silos. On one hand, the platforms do not share the original core sensor data, making it difficult for third-party developers or agricultural socialized service organizations to conduct in-depth AI model training and secondary development based on the underlying data. On the other hand, the closed ecosystem prevents drones from different brands from accessing a unified platform for cross-device scheduling, significantly limiting centralized management and data fusion for large-scale, multi-model collaborative operations.

[0007] (2) Traditional monitoring platforms based on pure software thresholds (poor adaptability to dynamic environments and high false alarm rate) Another type of commonly used third-party monitoring platform on the market relies heavily on static rules or fixed thresholds in pure software (such as triggering an alarm simply by judging that the GPS signal strength drops to a certain value) to determine anomalies. However, agricultural operating environments are highly non-stationary and dynamic (such as sudden signal blockage in hilly areas and airflow disturbances to sensors in severe weather). Traditional fixed threshold models lack the ability to dynamically adapt to environmental noise and the ability to extract deep features from multi-dimensional heterogeneous data (such as the coupling relationship between attitude, electrical quantity, and signal). This not only leads to a huge discrepancy between the detection model being effective in laboratory environments and failing in real-world field scenarios, but also results in extremely high false alarm and false negative rates in actual operations. This causes operator fatigue, rendering the early warning system ineffective in providing practical safety guidance.

[0008] In summary, existing solutions have significant shortcomings in data openness, cross-platform collaboration, and AI adaptive early warning capabilities in complex environments, making it difficult to meet the urgent needs of modern smart agriculture for highly reliable and safe monitoring of drones. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides an intelligent monitoring and multi-drone collaborative system for agricultural drones based on hardware simulation.

[0010] The present invention is implemented as follows: an intelligent monitoring and multi-drone collaborative system for agricultural drones, comprising a hardware simulation node, a serial port data access module, a hybrid anomaly detection module, a multi-drone fusion sensing module, a real-time communication module, a front-end display layer, and a data storage layer; The hardware simulation node is used to generate agricultural drone operation status data according to asynchronous timing logic, and encapsulate the operation status data into a standard data exchange format and send it to the serial port data access module via a physical serial port. The operation status data includes drone identification, global positioning data, signal strength, voltage, current and authorization status. The serial port data access module is used to listen to the byte stream reported by the physical serial port, decode the byte stream, deserialize it, and verify the validity of its fields. The module then inputs the verified running status data into the hybrid anomaly detection module. The hybrid anomaly detection module is used to execute an anomaly detection process according to a sequence. The anomaly detection process includes: forming a window sequence by extracting operational status data based on a sliding window; cleaning the window sequence with low-level noise according to the three-standard-deviation criterion; inputting the cleaned window sequence into a time-series prediction model to obtain the predicted flight status value at the next moment; generating an anomaly judgment result based on the error between the predicted flight status value at the next moment and the actual flight status value; and inputting the anomaly judgment result and a feature vector formed by position offset, velocity change rate, and signal attenuation rate into a classification model to output the anomaly type and risk level. The multi-drone fusion perception module is used to establish independent state buffers for different drones based on drone identifiers, and to perform multi-drone interference judgment and collaborative task status management based on the operating status data in each state buffer. The real-time communication module is used to push the anomaly type and risk level output by the hybrid anomaly detection module and the collaborative task status output by the multi-machine fusion perception module to the front-end display layer, and call the persistence interface to write the corresponding data into the data storage layer. The front-end display layer is used to update the drone real-time monitoring interface and 3D trajectory map based on the received data.

[0011] Furthermore, the hardware simulation node includes an embedded microcontroller unit, a firmware simulation unit, and a serial port output unit. The firmware simulation unit is used to simulate latitude and longitude drift, speed changes, signal attenuation, and power consumption during the flight of the agricultural drone. The serial port output unit is used to output the packaged operating status data to the physical serial port at a set transmission frequency.

[0012] Furthermore, the time-series prediction model in the hybrid anomaly detection module is a long short-term memory network model. The long short-term memory network model is used to capture the nonlinear variation trend of global positioning data and signal strength within a continuous window and output the predicted flight status value at the next moment.

[0013] Furthermore, the classification model is an isolated forest classification model, which is used to output anomaly types based on position offset, velocity change rate, and signal attenuation rate. The anomaly types include weak GPS signal, sudden position change, velocity anomaly, and signal loss. The risk levels include low risk, medium risk, and high risk.

[0014] Furthermore, the real-time communication module includes an asynchronous connection management unit and a message distribution unit. The asynchronous connection management unit is used to maintain a front-end client connection pool, and the message distribution unit is used to push the data in the corresponding status buffer to the matching front-end client according to the drone identifier, and write the same data stream into the data storage layer.

[0015] Another objective of this invention is to provide a method for generating and transmitting semi-physical simulation data of agricultural drones in a closed loop, comprising the following steps: simulating the flight state of a drone using an embedded microcontroller, dynamically generating five-dimensional state data through an asynchronous timed task, encapsulating the data into a JSON string, and transmitting it via a serial port at high frequency; deploying a listening service on a computer, establishing a long connection with the serial port of the embedded microcontroller, continuously receiving and reading the byte stream reported by the serial port, decoding, deserializing, and validating the original data; pushing the validated data stream to an asynchronous connection management center, performing dual-track parallel processing, one track distributing the data to the front-end presentation layer via the WebSocket protocol, and the other track synchronously calling a persistent interface to store the data in the data storage layer.

[0016] Furthermore, the embedded microcontroller employs firmware logic based on asynchronous timed tasks, dynamically simulating latitude and longitude drift, speed transients, signal attenuation, and linear battery power consumption through mathematical models and a random number generator.

[0017] Furthermore, the five-dimensional status data includes the drone's identifier, longitude, latitude, speed, and battery level.

[0018] Furthermore, the monitoring service establishes a long connection with the serial port using a serial communication library, continuously monitors and reads the byte stream reported by the hardware, and performs decoding and JSON deserialization.

[0019] Furthermore, historical flight data is stored in CSV or TXT file format. The data management and import module supports the import of flight data in CSV and TXT formats for post-flight analysis and simulation testing.

[0020] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows: (1) Core technological innovation of multi-source fusion × lightweight algorithm The core competitiveness of this invention lies in the innovative combination of multi-source fusion data and lightweight algorithms.

[0021] At the data level, it innovatively integrates heterogeneous data from multiple sources such as GPS positioning, flight speed, signal strength, and battery power to achieve multi-dimensional data fusion, making monitoring more comprehensive and accurate, and breaking through the limitations of traditional monitoring data being singular.

[0022] At the algorithmic level, it innovatively integrates a lightweight LSTM time series model, isolated forest unsupervised anomaly screening, and 3D model. A statistical noise reduction method is developed, relying on a sliding window to collect multi-dimensional operational characteristics of UAVs in real time. Combined with measured field data such as position offset, speed fluctuations, and signal attenuation, it completes anomaly identification and risk level classification. This constructs a detection system adapted to complex field operating environments, effectively addressing the practical pain points of traditional anomaly detection algorithms: weak generalization, poor adaptability to real-world conditions, and good laboratory results but difficult to implement in field scenarios. This enables real-time flight status analysis, rapid anomaly identification, and tiered early warning.

[0023] In terms of technical implementation, an innovative Web architecture is adopted, with Python supporting data processing and algorithm deployment on the backend, and Vue + ECharts framework combined with GIS map technology on the frontend to realize real-time data interaction and historical task playback, which greatly reduces the threshold for use by grassroots users and makes the technology lightweight and adaptable to ordinary terminal devices.

[0024] (2) Experience innovation that combines visualization and practicality The system visually presents monitoring results through GIS maps and status warning indicators, creating a dual effect of real-time early warning and post-event optimization, making the monitoring results visible, usable, and optimizable. The real-time early warning function ensures that users can detect anomalies immediately during operations, while the post-event optimization function supports users in reviewing and analyzing the process to optimize subsequent operational strategies.

[0025] (3) Precise adaptation and innovation in complex agricultural scenarios This system is deeply rooted in complex agricultural scenarios such as hilly and mountainous areas, and orchards and forests, solving the industry problem of general solutions failing in the field. The system integrates a real-time weather monitoring module to update meteorological data such as wind speed and precipitation in the operating area in real time, building an early warning system to predict adverse weather in advance. Combining anomaly analysis with real-time weather data, it provides professional and optimal suggestions for plant protection operations, truly achieving a closed loop from monitoring to decision-making.

[0026] (4) Semi-physical simulation technology based on physical serial port protocol mapping The project abandons the limitations of traditional monitoring systems that rely on static file reading, and innovatively constructs a semi-physical simulation environment that integrates hardware drivers, protocol hubs, and dynamic sensing.

[0027] Physical link established: The system uses a compatible Arduino UNOR3 to simulate the drone's flight attitude, sending byte streams in real time via USB serial port, and building a protocol conversion engine on the backend to map them into standard web data packets. This architecture verifies the system's ability to connect to a real data radio and directly take over the communication of the physical drone.

[0028] Practical algorithm verification: Under real physical communication links, extreme conditions such as signal delay and GPS packet loss are simulated using hardware clocks. This enables the backend 3 Statistical analysis and lightweight LSTM models underwent in-depth stress testing under non-ideal data sources, significantly improving the feasibility of the algorithm in plant protection field applications.

[0029] Advantages of engineering closed loop: Through the intervention of physical hardware, the project has formed a complete industrial-grade link of physical triggering - algorithm recognition - storage on disk - front-end early warning, making it not only a monitoring system, but also a universal communication framework that can be adapted to a variety of agricultural smart terminals.

[0030] Currently, the agricultural drone industry, both domestically and internationally, suffers from a technological blind spot: prioritizing flight control over signal security. Traditional flight control systems often rely on simple threshold judgments (such as satellite count and HDOP value) for GPS signal quality detection, failing to effectively identify signal anomalies unique to agriculture, such as those caused by hilly terrain obstruction, electromagnetic interference from high-voltage power lines, multipath effects, and spoofing attacks. While professional-grade spectrum analyzers or differential base stations offer high accuracy, their cost per unit ranges from tens of thousands to hundreds of thousands of yuan, making them impractical for low-profit plant protection service organizations. This invention employs a pure software architecture, integrating multi-source data from GPS, IMU, visual optical flow, and RSSI to construct a hybrid detection model combining statistical analysis and lightweight machine learning (LSTM, Isolation Forest). This model requires no additional hardware and can achieve millisecond-level anomaly identification on ordinary smartphones or low-computing-power chips integrated into flight controllers. It fills the industry gap in high-precision, low-cost, and readily deployable real-time signal monitoring technology, providing a universally applicable technological path for agricultural drones, moving from blind flight to early warning.

[0031] Existing academic research and industrial-grade anomaly detection solutions are mostly built on publicly available datasets in urban environments or aviation fields, and their generalization ability in scenarios such as farmland, orchards, and forests is severely insufficient. In particular, there has been a long-standing lack of systematic data accumulation and targeted algorithm research both domestically and internationally regarding long-tail risks such as intermittent signal obstruction in hilly terrain, electromagnetic field disturbances unique to high-voltage power lines, and deception interference designed for agricultural flight control protocols. Based on joint verification scenarios with cooperatives, our team independently collected and constructed a dedicated dataset containing over 100,000 real agricultural anomaly data points, each labeled with the environment type, cause of the anomaly, and severity level. On this basis, we invented an anomaly discrimination mechanism based on multi-source signal consistency, capable of distinguishing between natural drift caused by terrain obstruction and artificial bias caused by deception attacks, and generating a three-color risk warning (green-yellow-red) readable within seconds. This technical approach effectively solves the industry problem of high false alarm and false negative rates in complex agricultural environments, filling the gap in signal anomaly recognition technology specifically for agricultural scenarios.

[0032] Internationally, consumer and industrial drones generally employ their own closed communication protocols, resulting in incompatible flight control data formats across different brands and models, making it difficult to deploy third-party safety monitoring software across platforms. Another key technological innovation of this invention lies in the independently developed "Real-time Data Exchange Protocol" for agricultural scenarios. This protocol unifies the JSON format and communication timing of four key data types: GPS, IMU, RSSI, and visual optical flow, and has been submitted for software copyright. This protocol can be quickly adapted to mainstream agricultural drone models such as the DJI MG / T series and XAG XP series, and also supports reading flight control broadcast data via a mobile app bridge, significantly lowering the access barrier. Furthermore, the technical solution reserves an edge computing interface for embedding lightweight detection models into the flight control firmware, enabling the evolution from cloud monitoring to real-time edge alarms. This standardized and scalable technical architecture fills the gap in the domestic agricultural drone field, which lacks a unified signal safety monitoring interface standard and cross-brand adaptability, providing crucial infrastructure for the industry to move from post-incident statistics to proactive prevention. Attached Figure Description

[0033] Figure 1 A system module partitioning diagram provided for embodiments of the present invention; Figure 2 The system architecture diagram provided for embodiments of the present invention; Figure 3 This is the core logic of the real-time monitoring data panel provided in the embodiments of the present invention; Figure 4 The core logic of the trajectory map provided in the embodiments of the present invention; Figure 5 The map flashing indicator provided in the embodiments of the present invention; Figure 6 This is the core logic for multi-machine collaboration provided in the embodiments of the present invention; Figure 7 Database design diagram provided for embodiments of the present invention; Figure 8 This is a schematic diagram illustrating the hardware simulation data effect provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the first interface provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the login and registration interface provided in an embodiment of the present invention; Figure 11 This is a schematic diagram of a drone monitoring interface provided in an embodiment of the present invention; Figure 12 This is a schematic diagram of a data import interface provided in an embodiment of the present invention; Figure 13 This is a schematic diagram of a risk assessment interface provided in an embodiment of the present invention; Figure 14 This is a schematic diagram of the plant protection assessment interface provided in an embodiment of the present invention; Figure 15 This is a schematic diagram of the situation visualization interface provided in an embodiment of the present invention; Figure 16 This is a schematic diagram of the anomaly detection parameter interface of the system configuration interface provided in this embodiment of the invention; Figure 17 This is a schematic diagram of the data format configuration interface of the system configuration interface provided in an embodiment of the present invention; Figure 18 This is a schematic diagram of the user management interface of the system configuration interface provided in this embodiment of the invention; Figure 19 This invention provides a user guide, frequently asked questions, and an interface diagram showing the exception types for the Help Center. Figure 20 This is a schematic diagram of the data format of the help center interface provided in an embodiment of the present invention; Figure 21 This is a schematic diagram of the contact information interface of the help center provided in an embodiment of the present invention; Figure 22 The real-time monitoring effect diagram provided in the embodiment of the present invention; Figure 23 This is a diagram illustrating the effect of anomaly recognition operation provided in an embodiment of the present invention. Figure 24 This is a data import effect diagram provided in an embodiment of the present invention; Figure 25 This is a schematic diagram of the risk assessment interface for selecting and analyzing datasets provided in an embodiment of the present invention; Figure 26 This is a schematic diagram illustrating the start of the risk assessment interface provided in an embodiment of the present invention. Figure 27 This is a risk detection effect diagram provided in an embodiment of the present invention; Figure 28 This is a schematic diagram illustrating the process of viewing plant protection recommendations as provided in an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0035] The intelligent monitoring and multi-drone collaborative system for agricultural drones provided in this embodiment of the invention includes a hardware simulation node, a serial port data access module, a hybrid anomaly detection module, a multi-drone fusion sensing module, a real-time communication module, a front-end display layer, and a data storage layer. The hardware simulation node is used to generate agricultural drone operation status data according to asynchronous timing logic, and encapsulate the operation status data into a standard data exchange format and send it to the serial port data access module via a physical serial port. The operation status data includes drone identification, global positioning data, signal strength, voltage, current and authorization status. The serial port data access module is used to listen to the byte stream reported by the physical serial port, decode the byte stream, deserialize it, and verify the validity of its fields. The module then inputs the verified running status data into the hybrid anomaly detection module. The hybrid anomaly detection module is used to execute an anomaly detection process according to a sequence. The anomaly detection process includes: forming a window sequence by extracting operational status data based on a sliding window; cleaning the window sequence with low-level noise according to the three-standard-deviation criterion; inputting the cleaned window sequence into a time-series prediction model to obtain the predicted flight status value at the next moment; generating an anomaly judgment result based on the error between the predicted flight status value at the next moment and the actual flight status value; and inputting the anomaly judgment result and a feature vector formed by position offset, velocity change rate, and signal attenuation rate into a classification model to output the anomaly type and risk level. The multi-drone fusion perception module is used to establish independent state buffers for different drones based on drone identifiers, and to perform multi-drone interference judgment and collaborative task status management based on the operating status data in each state buffer. The real-time communication module is used to push the anomaly type and risk level output by the hybrid anomaly detection module and the collaborative task status output by the multi-machine fusion perception module to the front-end display layer, and call the persistence interface to write the corresponding data into the data storage layer. The front-end display layer is used to update the drone real-time monitoring interface and 3D trajectory map based on the received data.

[0036] The hardware simulation node includes an embedded microcontroller unit, a firmware simulation unit, and a serial port output unit. The firmware simulation unit is used to simulate latitude and longitude drift, speed changes, signal attenuation, and power consumption during the flight of the agricultural drone. The serial port output unit is used to output the packaged operating status data to the physical serial port at a set transmission frequency.

[0037] The temporal prediction model in the hybrid anomaly detection module is a long short-term memory network model. The long short-term memory network model is used to capture the nonlinear variation trend of global positioning data and signal strength within a continuous window and output the predicted flight status value at the next moment.

[0038] The classification model is an isolated forest classification model. The classification model is used to output anomaly types based on position offset, velocity change rate and signal attenuation rate. The anomaly types include weak GPS signal, sudden position change, velocity anomaly and signal loss. The risk levels include low risk, medium risk and high risk.

[0039] The real-time communication module includes an asynchronous connection management unit and a message distribution unit. The asynchronous connection management unit is used to maintain a front-end client connection pool, and the message distribution unit is used to push the data in the corresponding status buffer to the matching front-end client according to the drone identifier, and write the same data stream into the data storage layer.

[0040] Preferably, the agricultural drone intelligent monitoring and multi-drone collaborative system provided in this embodiment of the invention includes a front-end display layer, a back-end service layer, and a data storage layer. The front-end display layer is used to visualize and display real-time drone data and present 3D dynamic trajectories, and is configured with a WebSocket real-time communication link. The back-end service layer includes a WebSocket server, a hybrid anomaly detection module, a multi-drone fusion perception module, and a data management and import module. The hybrid anomaly detection module is configured with a heterogeneous data processing flow executed sequentially. This flow uses a sliding window mechanism to extract five-dimensional heterogeneous data, uses statistical criteria to perform low-level noise cleaning, uses a lightweight deep learning model to predict flight status, and performs anomaly classification and risk rating based on an isolated forest classifier. The back-end service layer pushes the processed drone data to the front-end display layer in real time through the WebSocket server and synchronously calls the persistence interface to write the data to the data storage layer.

[0041] The front-end display layer includes a 3D GIS trajectory map component, a real-time monitoring dashboard component, a three-color risk indicator component, and a multi-machine switching view component.

[0042] The hybrid anomaly detection module processes five-dimensional heterogeneous data including GPS data, signal strength, voltage, current, and authorization status.

[0043] The hybrid anomaly detection module comprises three processing stages. The first stage utilizes 3 The first stage performs low-level noise cleaning, the second stage uses a lightweight deep learning model to capture the nonlinear variation trend of position and signal strength, and the third stage uses an isolated forest classifier to combine position offset, velocity change rate and signal attenuation rate to output anomaly type and risk level.

[0044] The backend service layer includes a cross-protocol connection module. This module adopts the adapter design pattern to be compatible with three IoT communication protocols: UDP, TCP, and MQTT. It builds a unified data parsing interface to complete the reception and format standardization of raw data.

[0045] As a preferred embodiment, the agricultural drone semi-physical simulation data generation and closed-loop transmission method of the present invention includes the following steps: simulating the drone's flight state using an embedded microcontroller, dynamically generating five-dimensional state data through an asynchronous timed task, encapsulating the data into a JSON string, and transmitting it via a serial port at high frequency; deploying a listening service on a computer, establishing a long connection with the serial port of the embedded microcontroller, continuously receiving and reading the byte stream reported by the serial port, decoding, deserializing, and validating the original data; pushing the validated data stream to the asynchronous connection management center, performing dual-track parallel processing, one track distributing it to the front-end presentation layer via the WebSocket protocol, and the other track synchronously calling the persistent interface to store it in the data storage layer.

[0046] The embedded microcontroller uses firmware logic based on asynchronous timed tasks to dynamically simulate latitude and longitude drift, speed transients, signal attenuation, and linear battery power consumption through mathematical models and random number generators.

[0047] The five-dimensional status data includes the drone's identifier, longitude, latitude, speed, and battery level.

[0048] The monitoring service establishes a long connection with the serial port using a serial communication library, continuously listens for and reads the byte stream reported by the hardware, and performs decoding and JSON deserialization.

[0049] Historical flight data is stored in CSV or TXT file format. The data management and import module supports the import of flight data in CSV and TXT formats for post-flight analysis and simulation testing.

[0050] As a preferred embodiment of the present invention, an intelligent monitoring and multi-drone collaborative system for agricultural drones based on hardware simulation is provided. The system adopts a front-end and back-end separation architecture and is divided into three parts: a front-end display layer, a back-end service layer, and a data storage layer. (1) Front-end presentation layer: Based on the Vue.js framework, combined with ECharts chart library and GIS map technology, it realizes real-time data visualization, 3D dynamic trajectory map, alarm component and multi-machine switching view; the front-end establishes a real-time communication link with the back-end through WebSocket, receives real-time data pushed by drones, and completes functions such as historical data query and parameter configuration through HTTP requests; (2) Backend service layer: Built on Python and the Django / Flask framework, responsible for core business logic processing; mainly includes: WebSocket server to realize real-time data push; anomaly detection algorithm module, integrating LSTM, 3D... Statistical analysis models; multi-aircraft fusion perception logic to determine multi-aircraft interference; anomaly level assessment to output different risk levels; data management and import module to support the import of flight data in multiple formats such as CSV and TXT; (3) Data storage layer: MySQL database is used to store user information, flight mission data, anomaly records, system configuration, etc.; historical flight data is stored in CSV file format for simulation testing and post-event analysis.

[0051] The system constructs a hybrid anomaly detection model that integrates statistical analysis, machine learning, and deep learning. This architecture is designed to handle high-noise, non-stationary, heterogeneous data streams in complex agricultural environments and achieves accurate perception of the drone's operational status through multi-level filtering and recognition logic. The system employs a sliding window mechanism to extract five-dimensional heterogeneous data pushed via real-time physical serial ports, including GPS, signal strength, voltage, current, and authorization status. The detection process consists of three stages: First, the raw flight data is cleaned for noise and coarsely screened for anomalies using the 3σ criterion combined with an isolated forest unsupervised algorithm, eliminating outliers and identifying potential anomalies. Then, a lightweight LSTM time-series model is used to capture the nonlinear trends of data such as position, signal strength, and flight speed. By predicting the flight state at the next moment and comparing the reconstruction error between the predicted and actual values, accurate anomaly determination is achieved. Finally, based on the sliding window and multi-dimensional feature analysis, the anomalies identified by the LSTM are classified and risk-rated. Combining features such as position offset, velocity change rate, and signal attenuation rate, specific anomaly types such as weak GPS signal, sudden position change, and abnormal velocity are output, and a three-level risk assessment (low, medium, and high) is completed.

[0052] This system innovatively integrates multi-source heterogeneous data to achieve multi-dimensional data fusion. By fusing multi-source heterogeneous data such as GPS positioning, flight speed, signal strength, and battery level, the system can comprehensively judge the flight status from multiple dimensions, breaking through the limitations of traditional monitoring data with only one dimension. It utilizes 3D... Statistical analysis performs preliminary data cleaning on the speed, altitude, and position data transmitted by the drone, filtering out significant sensor noise. To address the random high-frequency noise generated by agricultural drones in complex field environments, the system establishes a dynamic sliding window, calculates the mean and standard deviation of the data within the window, and removes outliers through interval discrimination. This provides high-quality, smooth input features for subsequent deep learning models, effectively avoiding model fluctuations caused by outliers and laying the foundation for accurate detection.

[0053] The system introduces an unsupervised isolated forest algorithm to perform initial anomaly screening and quickly locate potential anomalous data segments. Subsequently, a lightweight LSTM time series model is used to capture the nonlinear changing trends of data such as position, signal strength, and flight speed, predict the flight state at the next moment, and calculate the prediction error to complete the accurate anomaly determination. Finally, based on multidimensional feature analysis and sliding window statistics, the anomalies identified by LSTM are classified and risk-rated. Combining features such as position offset, velocity change rate, and signal attenuation rate, specific anomaly types such as weak GPS signal, sudden position change, altitude anomaly, and velocity anomaly are output, and three risk levels of low, medium, and high are assessed.

[0054] The system is designed with a cross-protocol connection module to achieve compatibility and adaptation with three mainstream IoT communication protocols: UDP, TCP, and MQTT, and to build a unified data parsing interface; This module adopts the adapter design pattern, building independent parsing units for the three protocols to complete the reception of raw data, frame parsing, field verification and format standardization. The parsed output strictly follows the system's unified data exchange specifications and is completely consistent with the defined core fields, ensuring that the parsed data can be directly connected to the upper-layer anomaly identification, real-time visualization and data storage modules without the need for additional field conversion. The unified parsing interface has a built-in field validation mechanism that performs legality verification based on preset value ranges and data types, automatically filtering invalid data to ensure data integrity and accuracy. This design effectively improves the system's device compatibility and scalability, enabling multiple protocol data to be parsed once and reused across all modules, providing technical support for collaborative monitoring of multiple drones in complex agricultural operation scenarios.

[0055] This system employs a semi-physical simulation mechanism based on a compatible version of Arduino UNOR3 for data simulation. The complete data link connects the Arduino hardware node → USB serial port → Python backend hub → WebSocket real-time distribution → front-end visualization webpage. The specific implementation scheme is as follows: (1) Hardware simulation implementation: Arduino UNOR3 is used as the edge physical simulation node. The underlying firmware is written in C++ to simulate the state output of the UAV flight control system. In terms of code logic, the blocking delay() is abandoned and the millis() function is used to build an efficient asynchronous timed task. The mathematical model and random number generator are used to dynamically simulate the small latitude and longitude drift, speed transient, signal attenuation and linear consumption of battery power in flight operations. The system will strictly encapsulate the generated five-dimensional state data, drone_id, longitude, latitude, speed, battery_level, etc. into a standard JSON string and send it out at high frequency and stably via the USB physical serial port through the Serial.println() instruction. (2) Data transfer and protocol conversion layer implementation: A dedicated Python serial port monitoring service is deployed on the computer to serve as the hub for converting hardware-level physical protocols to Web network protocols. This service establishes a long connection with the COM port mounted on the Arduino using the pyserial library, and continuously listens for and reads the byte stream reported by the hardware through serial.readline(). After receiving the raw data, decode() is executed to decode and deserialize JSON, and the basic field validity is checked. The data stream that passes the check is then seamlessly pushed into the asynchronous connection management center of the system, which eventually triggers the dual-track parallel processing logic: one path is distributed to the 3D GIS map and dashboard of the multi-machine monitoring interface with a millisecond delay through the WebSocket protocol, and the other path synchronously calls the persistent interface to store it in the MySQL database, realizing a complete business loop from the physical hardware end to the digital application end.

[0056] The intelligent monitoring and multi-drone collaborative system for agricultural drones provided in this invention uses hardware simulation as the data starting point. The system utilizes an embedded microcontroller as an edge physical simulation node, and dynamically generates five-dimensional heterogeneous data, including GPS positioning, signal strength, voltage, current, and authorization status, through underlying firmware and asynchronous timing logic. This data is encapsulated in standard JSON format and transmitted to the backend central control via a high-frequency USB physical serial port.

[0057] The backend service layer listens for and receives the byte stream reported via the serial port through a serial communication library, performing decoding, deserialization, and field validation on the raw data. For multi-source heterogeneous data, the system constructs a hybrid anomaly detection model integrating statistical analysis, machine learning, and deep learning. This model first uses a sliding window mechanism to truncate the data, combined with 3... The criteria involve low-level noise cleaning to filter high-frequency noise from the sensor under complex airflow conditions; then, a lightweight LSTM model is used to capture the nonlinear variation trends of data such as position and signal strength; finally, a classifier based on the isolated forest algorithm is used, combined with features such as position offset and velocity change rate, to classify the anomalies identified by LSTM into risk levels, accurately outputting the anomaly type and low, medium, or high risk level.

[0058] After the data is processed by the algorithm engine, the backend WebSocket server is responsible for pushing the processed real-time data to the frontend presentation layer with millisecond-level latency. Simultaneously, the backend service layer incorporates multi-aircraft fusion sensing logic to achieve multi-aircraft interference detection and collaborative task management. Historical flight data supports importing and storing in CSV and TXT formats. The system synchronously writes the data to the data storage layer through a persistent interface, ensuring the traceability of flight records.

[0059] The front-end is built on the Vue.js framework. After receiving real-time data via the WebSocket protocol, it triggers core methods to update data objects, synchronously refreshing dashboard components such as signal strength, battery level, GPS satellite count, and flight speed. For 3D scenes, the front-end GIS engine adopts a dynamic rendering strategy, prioritizing the existence of the drone entity; if it already exists, its position coordinates are updated directly; if it is a new drone, the new entity is dynamically added, its 3D model is loaded, and an ID tag is bound, enabling real-time updates of multi-drone view switching and 3D trajectory maps. The system thus fully integrates the entire closed-loop process from physical simulation, data analysis, intelligent algorithms to visual interaction.

[0060] like Figure 1 , 2 As shown, this embodiment of the invention provides an intelligent monitoring and multi-drone collaborative system for agricultural drones based on hardware simulation. The system can be divided into multiple modules such as data access, anomaly detection, visualization display, and user management. The overall system adopts a front-end and back-end separation architecture, which is divided into three parts: front-end display layer, back-end service layer, and data storage layer. (1) Front-end presentation layer: Based on the Vue.js framework, combined with ECharts chart library and GIS map technology, it realizes real-time data visualization, 3D dynamic trajectory map, alarm component and multi-machine switching view; the front-end establishes a real-time communication link with the back-end through WebSocket, receives real-time data pushed by drones, and completes functions such as historical data query and parameter configuration through HTTP requests; (2) Backend service layer: Built on Python and the Django / Flask framework, responsible for core business logic processing; mainly includes: WebSocket server to realize real-time data push; anomaly detection algorithm module, integrating LSTM, 3D... Statistical analysis models; multi-aircraft fusion perception logic to determine multi-aircraft interference; anomaly level assessment to output different risk levels; data management and import module to support the import of flight data in multiple formats such as CSV and TXT; (3) Data storage layer: MySQL database is used to store user information, flight mission data, anomaly records, system configuration, etc.; historical flight data is stored in CSV file format for simulation testing and post-event analysis.

[0061] The system constructs a hybrid anomaly detection model that integrates statistical analysis, machine learning, and deep learning. This architecture is designed to handle high-noise, non-stationary, heterogeneous data streams in complex agricultural environments and achieves accurate perception of the drone's operational status through multi-level filtering and recognition logic. The system employs a sliding window mechanism to capture five-dimensional heterogeneous data pushed in real-time via the physical serial port, including GPS, signal strength, voltage, current, and authorization status. The detection process is divided into three stages: firstly, using 3D... The criteria are to perform low-level noise cleanup; then, a lightweight LSTM model is used to capture the nonlinear changing trends of data such as position and signal strength to predict the flight status at the next moment. By comparing the error between the predicted value and the actual value, it is determined whether there is an anomaly; finally, the anomaly identified by the LSTM model is further classified and risk-rated by the machine forest classifier. Combining the input features such as position offset, velocity change rate, and signal attenuation rate, the specific anomaly type (such as weak GPS signal, sudden position change, etc.) and low / medium / high risk level are output.

[0062] This system innovatively integrates multi-source heterogeneous data to achieve multi-dimensional data fusion. By fusing data from multiple sources such as longitude, latitude, speed, signal strength, GPS satellite count, and battery level, the system can comprehensively assess flight status from multiple dimensions, overcoming the limitations of traditional single-source monitoring data. It utilizes 3D... Statistical analysis performs preliminary data cleaning on linear indicators such as speed and altitude transmitted by the drone, filtering out obvious sensor noise. For the random high-frequency noise generated by agricultural drone sensors under complex airflow conditions, the system first establishes a dynamic sliding window and calculates the mean of the data within the window. ) and standard deviation ), by judging the interval [ -3 , +3 This method removes outliers, providing high-quality, smooth input features for subsequent deep learning models. It effectively avoids fluctuations in model predictions caused by a single outlier, laying the foundation for subsequent model detection.

[0063] The system introduces an unsupervised isolated forest algorithm to complete the initial screening of abnormal data; then, a lightweight LSTM time series model is used to capture nonlinear features and predict the state of flight data to achieve accurate anomaly identification; finally, based on sliding window and multidimensional feature statistical analysis, the anomalies identified by the LSTM model are classified and risk-rated. Combining features such as position offset, velocity change rate, and signal attenuation rate, specific anomaly types such as weak GPS signal, sudden position change, and abnormal velocity are output, and risk assessments of low, medium, and high levels are completed.

[0064] Depending on the severity of the anomaly, the system outputs different risk levels: Table 1. Definition of UAV Anomaly Levels and Handling Recommendations

[0065] The system is designed with a cross-protocol connection module to achieve compatibility and adaptation with three mainstream IoT communication protocols: UDP, TCP, and MQTT, and to build a unified data parsing interface; This module adopts the adapter design pattern, building independent parsing units for the three protocols to complete the reception of raw data, frame parsing, field verification and format standardization. The parsed output strictly follows the system's unified data exchange specifications and is completely consistent with the defined core fields, ensuring that the parsed data can be directly connected to the upper-layer anomaly identification, real-time visualization and data storage modules without the need for additional field conversion. The unified parsing interface has a built-in field validation mechanism that performs legality verification based on preset value ranges and data types, automatically filtering invalid data to ensure data integrity and accuracy. This design effectively improves the system's device compatibility and scalability, enabling multiple protocol data to be parsed once and reused across all modules, providing technical support for collaborative monitoring of multiple drones in complex agricultural operation scenarios.

[0066] This system employs a semi-physical simulation mechanism based on a compatible version of Arduino UNOR3 for data simulation. The complete data link connects the Arduino hardware node → USB serial port → Python backend hub → WebSocket real-time distribution → front-end visualization webpage. The specific implementation scheme is as follows: (1) Hardware simulation implementation: Arduino UNOR3 is used as the edge physical simulation node. The underlying firmware is written in C++ to simulate the state output of the UAV flight control system. In terms of code logic, the blocking delay() is abandoned and the millis() function is used to build an efficient asynchronous timed task. The mathematical model and random number generator are used to dynamically simulate the small latitude and longitude drift, speed transient, signal attenuation and linear consumption of battery power in flight operations. The system will strictly encapsulate the generated five-dimensional state data, drone_id, longitude, latitude, speed, battery_level, etc. into a standard JSON string and send it out at high frequency and stably via the USB physical serial port through the Serial.println() instruction. (2) Data transfer and protocol conversion layer implementation: A dedicated Python serial port monitoring service is deployed on the computer to serve as the hub for converting hardware-level physical protocols to Web network protocols. This service establishes a long connection with the COM port mounted on the Arduino using the pyserial library, and continuously listens for and reads the byte stream reported by the hardware through serial.readline(). After receiving the raw data, decode() is executed to decode and deserialize JSON, and the basic field validity is checked. The data stream that passes the check is then seamlessly pushed into the asynchronous connection management center of the system, which eventually triggers the dual-track parallel processing logic: one path is distributed to the 3D GIS map and dashboard of the multi-machine monitoring interface with a millisecond delay through the WebSocket protocol, and the other path synchronously calls the persistent interface to store it in the MySQL database, realizing a complete business loop from the physical hardware end to the digital application end.

[0067] Table 2 Hardware Simulation and Core Logic of Serial Port Forwarding (Arduino C++ & Python Relay)

[0068] Another objective of this invention is to provide an information data processing terminal, which is used to implement the hardware simulation-based intelligent monitoring and multi-drone collaborative system for agricultural drones.

[0069] Evidence related to the technical effects obtained by the embodiments of the present invention.

[0070] Backend and Communication Engineering: In multi-machine collaborative monitoring scenarios, real-time performance is a core prerequisite for security early warning. The system uses websocket__manager.py to build an asynchronous connection management center, achieving second-level data push based on the WebSocket protocol. Furthermore, specific performance optimizations have been performed for multi-machine concurrent scenarios, resolving clock synchronization and data congestion issues in multi-machine interference detection. The specific implementation is as follows: Backend Implementation: A WebSocket service is built using Python's websockets library combined with the asynchronous features of FastAPI, avoiding the performance bottlenecks of traditional synchronous services. A message processing function is defined in websocket__manager.py to receive data pushed via hardware simulation. After algorithmic processing, an independent state buffer is maintained for each online drone, and message streams are automatically distributed based on the drone__id to ensure no data confusion during multi-drone concurrency. Simultaneously, a multi-client connection pool is maintained to support precise push of real-time data from the corresponding drone to different clients, avoiding invalid data transmission. Furthermore, in terms of processing logic, the system adopts a dual-track parallel data flow mechanism: on one hand, after data processing, it enters the independent state buffer of each online drone, and message streams are automatically distributed based on the drone__id and precisely pushed to the client connection pool; on the other hand, the system synchronously / asynchronously triggers the data persistence interface to write the real-time parsed data to the database storage module. This ensures the synchronization between real-time dynamic display and historical trajectory storage, avoiding data confusion in multi-drone concurrency scenarios.

[0071] Front-end implementation: In the Vue component MultiDroneMonitor.vue, a WebSocket connection is created, and an onmessage callback function is defined to update the interface in real time after receiving data pushed from the backend; a built-in automatic reconnection mechanism is provided, along with a heartbeat detection function, to ensure connection stability in extreme network environments and avoid warning delays caused by connection interruptions.

[0072] Data format: In accordance with the Real-time Data Exchange Protocol, JSON format is used for transmission, simplifying data fields and compressing data volume to further improve push efficiency and ensure that the front end can still achieve zero-latency rendering of the trajectory under high concurrency.

[0073] Front-end and GIS Engineering: 1. Real-time monitoring data panel implementation The real-time monitoring data panel uses ECharts charts to display information such as signal strength, battery level, and GPS satellite count transmitted back by the drone in real time.

[0074] Implementation: A data panel is encapsulated using Vue components; real-time data is received via WebSocket to update the Vue reactive data; an ECharts chart instance listens for data changes and calls the setOption method to update the chart. This provides the core code logic for real-time monitoring of the data panel, such as... Figure 5 As shown.

[0075] 3D dynamic trajectory map implementation The 3D dynamic trajectory map is based on GIS map technology. The drone icon moves in real time with the data, and the trajectory line is dynamically drawn. The core goal is to adapt to large-scale agricultural operation scenarios, achieve accurate matching of latitude, longitude and altitude data, and restore the flight status under complex terrain.

[0076] Technology Selection: The system underwent a targeted evaluation for its rendering engine, ultimately selecting Cesium.js as the core rendering engine compared to the more common Three.js. This decision aligns with the actual needs of agricultural operations: Cesium natively supports the WGS84 global geographic coordinate system, enabling precise alignment of latitude, longitude, and altitude data transmitted from drones, avoiding coordinate drift issues in large-scale scenes such as hilly areas and vast farmlands. Simultaneously, the system integrates MapboxGLJS vector base maps, allowing for smooth loading of high-resolution satellite imagery in low-bandwidth field environments, clearly reproducing complex terrain features such as hills and forest networks, and providing precise support for the spatial positioning of anomalies.

[0077] Functionality includes: real-time drone icon location updates; historical trajectory line drawing (supporting different colors to distinguish normal / abnormal segments); abnormal point marking; camera view tracking; and 2D / 3D view switching. Its core logic is as follows: Figure 6 As shown.

[0078] 2. Real-time alarm component implementation The real-time alarm component features a prominent alarm UI. Upon receiving an anomaly, a warning dialog box immediately pops up on the screen, and a flashing marker appears at the corresponding location on the map. Its core logic is as follows: Figure 7 As shown.

[0079] Alarm component design: Use Vue's transition animations to implement the pop-up effect; use different colors according to the risk level.

[0080] Alarm information includes: drone ID, anomaly type, risk level, and suggested handling measures; manual and automatic shutdown are supported.

[0081] 3. Multi-machine view switching implementation The multi-aircraft switching view design interaction function allows users to click on an aircraft icon, and the sidebar will immediately switch to display the detailed real-time status of that aircraft.

[0082] (1) Implementation scheme: The drone list component displays all online drones and supports clicking to select them; the drone icon on the map supports clicking to select; after selection, the sidebar displays the drone's detailed information (speed, altitude, signal strength, battery level, etc.); the camera view can follow the selected drone.

[0083] Table 3. Pseudocode for Algorithm1 Multi-Machine Switching View

[0084] 4. Implementation of Anomaly Recognition and Detection Algorithms Sliding window data processing. The algorithm needs to process sliding window data transmitted via a protocol. The system maintains a fixed-size first-in-first-out (FIFO) data queue for each UAV as a sliding window, and the window size can be flexibly configured according to the operational scenario. Upon receiving a new data point, the system adds it to the tail of the corresponding UAV's queue while automatically removing older data from the head of the queue, ensuring that the window always retains the most recent continuous flight state sequence. Subsequently, the system performs multi-dimensional feature extraction on the data within the window and inputs these features into the subsequent anomaly detection model. This effectively captures the temporal dependencies of flight states, avoids misjudgments caused by single-point data fluctuations, and controls computational overhead by limiting the window size, thereby achieving real-time anomaly detection and accurate risk warning for UAV signals.

[0085] Table 4 Core Logic of Sliding Window Feature Extraction

[0086] Multi-drone fusion perception logic. The multi-drone interference judgment logic is achieved by analyzing the relative positions and flight parameters of multiple drones, and makes a comprehensive judgment from three dimensions: First, spatial proximity detection, which calculates the three-dimensional Euclidean distance between each pair of drones. When the distance is less than 30 meters, a high-risk close proximity warning is triggered, and when it is between 30 and 80 meters, a medium-risk warning is triggered. Second, collision risk prediction, which calculates the approach rate by combining velocity vector and heading angle. When two drones are moving towards each other and the approach rate is greater than 5 m / s, the estimated time of collision (TTC) is output and a high-risk collision risk warning is triggered. Third, signal interference detection, which triggers a medium-risk signal interference warning when the distance between two drones is less than 150 meters and the signal strength of both drones is below -75dBm and shows a synchronous decreasing trend, indicating that there may be co-channel interference.

[0087] Table 5 Core Logic of Multi-Machine Interference Detection Algorithm

[0088] Anomaly Level Assessment. Based on the severity of the anomaly, the system outputs different risk levels: Table 6. Core Logic of Risk Level Assessment Algorithm

[0089] Database design. The database design revolves around user management, flight data management, and anomaly logging, and mainly includes the following core components.

[0090] User Permissions Related Table user: This table stores basic user information, including username, password (encrypted), email address, and role. role: Role table, defines different user roles (administrator, regular user). permission: The permission table defines system operation permissions. user__role: User-role association table Flight data related tables drone__info: A table containing basic drone information, including drone ID, model, and owner. flight__task: Flight task table, storing task name, start time, end time, task status, etc. `flight__data`: This table stores real-time flight data, including timestamps, longitude, latitude, altitude, speed, signal strength, battery level, GPS satellite count, etc. data__import__record: A data import record table that records historical data import information. Exception Record Related Tables anomaly__record: An anomaly record table that stores the time, location, anomaly type, risk level, associated drone ID, and flight mission ID of the anomaly. anomaly__type: A dictionary of exception types, defining various exceptions and their descriptions. System configuration related tables system__config: The system configuration table stores configuration information such as anomaly detection parameters and early warning thresholds. The database core entity design relational model, such as... Figure 9 As shown.

[0091] UI Interface Demonstration User Interface Design: UI design is a crucial component of user experience, directly impacting the user's experience of using the system. This system provides a web-based user interface, primarily featuring the following interfaces: Based on actual usage needs, the project implemented a complete web-based user interface.

[0092] Homepage: Displays the main functions of the system.

[0093] Login screen: Enter your account and password for identity verification. It supports the "remember password" function and provides a registration redirect.

[0094] Registration interface: Fill in your username, email, password and other information to complete account registration.

[0095] Workbench Interface: Integrates a real-time weather workbench, interface shortcuts, and task display. It displays the current work area's wind speed, temperature, humidity, and weather conditions in a concise card format, and supports manual refresh.

[0096] Multi-UAV Monitoring Interface: The core system interface, integrating a status overview, a real-time map display area, and a collaborative early warning center. The map area uses GIS technology to display the real-time location and trajectory of the UAVs; the status overview bar displays real-time data such as the signal strength, battery level, and GPS satellite count of the currently selected UAV; the real-time map monitoring panel visualizes the real-time flight path, problems, and early warnings of the UAVs, supporting 2D and 3D conversion; the collaborative early warning center displays the real-time location of abnormal UAVs.

[0097] Data import interface: Supports uploading flight data files in formats such as CSV and TXT. After receiving the raw data uploaded by the user, it synchronously pushes it to the backend for verification and standardization processing, providing data support for the anomaly detection algorithm.

[0098] Risk assessment interface: After the user selects the imported dataset and adjusts the detection parameters, the backend anomaly detection algorithm is triggered. The detection results output by the algorithm are received, the anomaly locations and risk levels are displayed synchronously, and the user is redirected to the plant protection risk assessment details interface to obtain operational suggestions.

[0099] The plant protection risk assessment details interface includes basic information such as the assessment task name, time, region, detection algorithm, comprehensive risk level, plant protection management recommendations, system description, and anomaly details.

[0100] Situational visualization interface: Integrates flight trajectory visualization and an outlier list. Users can select the assessment task name, perform flight trajectory visualization and tracing, and analyze speed and altitude trends; the outlier list displays time, longitude, latitude, anomaly type, confidence level, and location information.

[0101] Results management interface: allows users to query historical flight missions, view historical trajectory replays, anomaly records, and analysis reports, and supports downloading.

[0102] System configuration interface: Supports configuration of anomaly detection parameters, including sliding window size, anomaly threshold, and warning threshold.

[0103] Help Center: Includes user tutorials, analysis of common problems, definitions and handling suggestions for exception types, data file specifications, and team contact information.

[0104] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0105] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent monitoring and multi-drone collaborative system for agricultural unmanned aerial vehicles (UAVs), characterized in that, It includes a hardware simulation node, a serial port data access module, a hybrid anomaly detection module, a multi-machine fusion sensing module, a real-time communication module, a front-end display layer, and a data storage layer; The hardware simulation node is used to generate agricultural drone operation status data according to asynchronous timing logic, and encapsulate the operation status data into a standard data exchange format and send it to the serial port data access module via a physical serial port. The operation status data includes drone identification, global positioning data, signal strength, voltage, current and authorization status. The serial port data access module is used to listen to the byte stream reported by the physical serial port, decode the byte stream, deserialize it, and verify the validity of its fields. The module then inputs the verified running status data into the hybrid anomaly detection module. The hybrid anomaly detection module is used to execute an anomaly detection process according to a sequence. The anomaly detection process includes: forming a window sequence by extracting operational status data based on a sliding window; cleaning the window sequence with low-level noise according to the three-standard-deviation criterion; inputting the cleaned window sequence into a time-series prediction model to obtain the predicted flight status value at the next moment; generating an anomaly judgment result based on the error between the predicted flight status value at the next moment and the actual flight status value; and inputting the anomaly judgment result and a feature vector formed by position offset, velocity change rate, and signal attenuation rate into a classification model to output the anomaly type and risk level. The multi-drone fusion perception module is used to establish independent state buffers for different drones based on drone identifiers, and to perform multi-drone interference judgment and collaborative task status management based on the operating status data in each state buffer. The real-time communication module is used to push the anomaly type and risk level output by the hybrid anomaly detection module and the collaborative task status output by the multi-machine fusion perception module to the front-end display layer, and call the persistence interface to write the corresponding data into the data storage layer. The front-end display layer is used to update the drone real-time monitoring interface and 3D trajectory map based on the received data.

2. The agricultural unmanned aerial vehicle (UAV) intelligent monitoring and multi-UAV collaborative system according to claim 1, characterized in that, The hardware simulation node includes an embedded microcontroller unit, a firmware simulation unit, and a serial port output unit. The firmware simulation unit is used to simulate latitude and longitude drift, speed changes, signal attenuation, and power consumption during the flight of the agricultural drone. The serial port output unit is used to output the packaged operating status data to the physical serial port at a set transmission frequency.

3. The intelligent monitoring and multi-drone collaborative system for agricultural unmanned aerial vehicles according to claim 1, characterized in that, The temporal prediction model in the hybrid anomaly detection module is a long short-term memory network model. The long short-term memory network model is used to capture the nonlinear variation trend of global positioning data and signal strength within a continuous window and output the predicted flight status value at the next moment.

4. The intelligent monitoring and multi-drone collaborative system for agricultural unmanned aerial vehicles according to claim 1, characterized in that, The classification model is an isolated forest classification model. The classification model is used to output anomaly types based on position offset, velocity change rate and signal attenuation rate. The anomaly types include weak GPS signal, sudden position change, velocity anomaly and signal loss. The risk levels include low risk, medium risk and high risk.

5. The intelligent monitoring and multi-drone collaborative system for agricultural unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The real-time communication module includes an asynchronous connection management unit and a message distribution unit. The asynchronous connection management unit is used to maintain a front-end client connection pool, and the message distribution unit is used to push the data in the corresponding status buffer to the matching front-end client according to the drone identifier, and write the same data stream into the data storage layer.

6. A method for generating and transmitting semi-physical simulation data of agricultural UAVs using the system described in any one of claims 1 to 5, characterized in that... Includes the following steps: An embedded microcontroller is used to simulate the flight state of a drone. Five-dimensional state data is dynamically generated through asynchronous timed tasks, and the data is encapsulated into a JSON string and sent out at high frequency via a serial port. A listening service is deployed on the computer to establish a long connection with the serial port of the embedded microcontroller, continuously receiving and reading the byte stream reported by the serial port, decoding, deserializing and validating the original data. The data stream that passes the validation is pushed to the asynchronous connection management center for dual-track parallel processing. One track distributes the data to the front-end presentation layer via the WebSocket protocol, while the other track synchronously calls the persistent interface to store the data in the data storage layer.

7. The method according to claim 6, characterized in that... The embedded microcontroller uses firmware logic based on asynchronous timed tasks to dynamically simulate latitude and longitude drift, speed transients, signal attenuation, and linear battery power consumption through mathematical models and random number generators.

8. The method according to claim 6, characterized in that... The five-dimensional status data includes the drone's identifier, longitude, latitude, speed, and battery level.

9. The method according to claim 6, characterized in that... The monitoring service establishes a long connection with the serial port using a serial communication library, continuously listens for and reads the byte stream reported by the hardware, and performs decoding and JSON deserialization.

10. The method according to claim 6, characterized in that... Historical flight data is stored in CSV or TXT file format. The data management and import module supports the import of flight data in CSV and TXT formats for post-flight analysis and simulation testing.

Images